Thumbnail image generated by Bing Image Creator, edited in Canva

Introduction

AI is omnipresent in the current technological landscape. Numerous articles highlight the latest functionalities of chat models like GPT or BERT, as well as intriguing stories of individuals employing multiple AIs simultaneously for various purposes. The emergence of novel roles in this domain is evident, with the most recent addition being that of a "prompt engineer."

Curious about the term "prompt engineer," I turned to Google's Bard for insights:

A prompt engineer is a specialized role in the field of artificial intelligence (AI) that focuses on designing and refining prompts for large language models (LLMs) and other AI models. Prompts are essentially instructions or questions that guide the AI model towards generating desired outputs. Prompt engineers play a crucial role in bridging the gap between human intentions and machine understanding."
[...]
-- Google Bard

It strikes me as amusing that there are already training programs for becoming a prompt engineer, considering the relatively nascent nature of this field. Predicting the future of such a role might be precarious, given the rapid evolution of generative AI models. It's plausible that this role might become obsolete in a few years, given the swift advancements in the capability of these models. Their existing limitations could well be overcome in a remarkably short span.

However, this doesn't imply that one should disregard this new technology. On the contrary, it's crucial to stay engaged with it, experiment, and incorporate it into our daily lives. Generative AI is already an integral part of our existence, and being prepared for its widespread integration is imperative. Don't miss the bus, or you risk to regret it.

What Are They Good At?

I've experimented with several AI platforms, including:

• ChatGPT
• Google Bard
• Microsoft Bing
• Bing Image Creator
• Github Copilot

My experiences with each vary:

ChatGPT:
Remains my favorite. Falls short for coding-related tasks, often producing hallucinated code that rarely works seamlessly. However, excels in writing, English correction, and enhancing overall text quality. Ideal for crafting articles, allowing me to achieve a language proficiency beyond my expectations. It proves effective in transforming notes into coherent articles. Notably inventive—my wife even uses it for exploring new book ideas. A downside is its tendency to adopt an overly enthusiastic "Apple touch," using excessive superlatives when a more straightforward description is needed. It's an article not a Keynote! Changing this tone is challenging.

Google Bard:
Limited usage due to inconclusive results. Code output was almost acceptable but exhibited some hallucinations. Less adept at bridging information gaps when provided with notes. Lacks inventiveness, often proposing commonplace and déjà vu ideas. Overall, my experience left me disappointed.

Microsoft Bing:
My experience with Microsoft Bing mirrors that of Bard. I found it to be quite useful in locating websites related to specific subjects and delivering concise summaries. Notably, it proved beneficial when I tasked it with evaluating javascript frontend frameworks based on a scale I defined. The results were satisfactory, showcasing its effectiveness in this context. However, similar to Bard, it falls short in terms of inventiveness.

Github Copilot:
Interesting but less applicable for my role as a Tech Lead in Data. Primarily focused on Data Engineering tasks involving Terraform and SQL development. Copilot's utility is limited for repetitive Terraform tasks, where a previously presented VSCode extension proves more beneficial. In SQL, it struggles to comprehend my intentions without knowledge of my data sources, their relations, formats, and field meanings. However, for JavaScript/TypeScript coding with clear function names, it efficiently proposes interesting code. Yet, due to its training on public GitHub repositories, it may suggest impractical or nonsensical solutions.

Bing Image Creator:
Utilized for creating thumbnail pictures for my posts. Produces pleasing results but has limitations: token restrictions, difficulty specifying a precise style, and an inability to adjust the picture format (size). Despite these drawbacks, it remains a valuable tool for someone lacking artistic capabilities, and I am generally satisfied with the outcomes.

Case Studies

I'll now delve into specific use cases to illustrate how I leverage generative AI:

Language Correction

My primary use is to "fix my English." As a native French speaker who grew up in the southern part of Belgium, my exposure to English was slightly higher than that of typical French speakers. However, like many, I face the challenge of maintaining a proficient command of English. Despite reading tech articles, following news, watching movies, and using English extensively online, my English is influenced by French, and my vocabulary is somewhat restricted.

When attempting to write, I find myself relying on a limited set of words, and my sentences bear a distinctively French flavour. Although my text is comprehensible, it often appears peculiar and lacks the natural flow expected by native speakers.

Here's a very simple example:

Prompt:

Fix and improve my English:
With generative AI like ChatGPT, I'm able to achieve much better readability.

Answer:

With generative AI, such as ChatGPT, I can significantly enhance readability.

Isn't that better?

Especially in large international corporations, where many communicate in "globish", generative AI has the potential to greatly elevate the quality of our communication.

Getting a Short Explanation About a Specific Subject

I frequently employ generative AI to quickly obtain concise explanations on specific subjects. An instance of this is evident at the beginning of this article:

Prompt:

What is a prompt engineer?

Answer:

“A prompt engineer is a specialized role in the field of artificial intelligence (AI) that focuses on designing and refining prompts for large language models (LLMs) and other AI models. Prompts are essentially instructions or questions that guide the AI model towards generating desired outputs. Prompt engineers play a crucial role in bridging the gap between human intentions and machine understanding.

The responsibilities of a prompt engineer typically include:

Understanding the capabilities and limitations of AI models: Prompt engineers need to have a deep understanding of the specific AI models they are working with, in order to effectively design prompts that will elicit the desired responses. This includes understanding the models' strengths and weaknesses, as well as the types of data and tasks they are trained on.

Designing effective prompts: Prompt engineers need to be able to design prompts that are clear, concise, and unambiguous. They also need to consider the context of the task at hand, and tailor the prompts accordingly. For example, a prompt for a summarization task would be different from a prompt for a creative writing task.

Evaluating and refining prompts: Prompt engineers need to be able to evaluate the effectiveness of their prompts, and make adjustments as needed. This may involve iteratively testing different prompts, and analyzing the results to see which ones produce the most desired outputs.

Collaborating with other stakeholders: Prompt engineers often work collaboratively with other stakeholders, such as data scientists, machine learning engineers, and product managers. This collaboration is essential for ensuring that prompts are aligned with the overall goals of an AI project.

The skills and qualifications required for a prompt engineer role typically include:
Strong understanding of natural language processing (NLP): Prompt engineers need to have a strong understanding of how language works, in order to design prompts that are effective at communicating with AI models.

Familiarity with AI models and machine learning: Prompt engineers need to be familiar with the different types of AI models and machine learning techniques, in order to understand how prompts can be used to influence their behavior.

Excellent communication and writing skills: Prompt engineers need to be able to communicate effectively with both technical and non-technical audiences. They also need to be able to write clear, concise, and well-structured prompts.

Problem-solving and analytical skills: Prompt engineers need to be able to identify problems with AI model outputs, and devise solutions in the form of refined prompts.

As AI technology continues to evolve, the role of prompt engineers is becoming increasingly important. Prompt engineers are essential for ensuring that AI models are able to understand and respond to human intentions in a meaningful way.
”
--Google Bard

It's comprehensive enough for me to quickly understand what this position is about. I don't need to search for an article on the internet, which might display numerous ads, be incomplete, or be outdated. If I wish to delve deeper into the subject, I'll then explore specialized websites.
Bing could be a perfect tool for this.

Translation

Whether it's a document or a single sentence, generative AI produces notably improved results compared to Google Translate, often enriching the output with additional context. For example, I'm presently immersed in a wood furniture project and employing a tool known as a "Trusquin."

Prompt:

Translate the french word “Trusquin”

Answer:

Bard - It knows french !

Sentence translations are equally effective, and the ability to request additional information enhances the overall utility. For instance:

Prompt:

Translate this french sentence in English, 
include a short description of what a queue d’aronde is in the translation, inline with the text:
Je travaille actuellement sur un projet de meuble télé en bois. 
J’ai envie de créer un beau meuble, sans vis, en utilisant la technique des queues d’aronde.

Answer:

I am currently working on a wooden TV stand project. I aim to craft an aesthetically pleasing piece of furniture without the use of screws, employing the dovetail technique. The dovetail technique, or "queue d'aronde" in French, involves creating trapezoidal-shaped projections (tails) on one piece of wood that interlock with corresponding notches (pins) on another piece. This technique results in a robust and visually appealing connection, enhancing the overall quality of the furniture.

Once again, the high quality of the generated text is evident. Trained on an extensive corpus of text, these AIs have access to a remarkable variety of subjects.

Woodworking techniques might not be the most intuitive topic to inquire about with an AI, yet it showcases the AI's proficiency in unexpected subjects:

Prompt:

What are the tools I’d need to make clean and precise dovetails ?

Answer:

ChatGPT - AI is not just about IT !

Writing Official Documents

Generative AI proves invaluable in crafting various official documents, whether it's your resume, a complaint letter to a contractor, or a letter to your local municipality. A friend of mine faced a parking predicament. Although his residence falls within a "free parking zone," being the last street before the paid area meant no available spots in his street due to others seeking free parking. He aimed to request a “resident card" for convenient parking. Turning to an AI, we effortlessly obtained a meticulously crafted letter, complete with all the necessary placeholders for addresses, names, and more. It's the kind of letter that would take hours to write, meticulously weighing every word. With an AI, it's done in minutes, eliminating the stress.

Writing an Article

Now, we delve into the 'inception' segment of this article: my use of ChatGPT to enhance my Hive articles. One might consider it a shortcut, assuming it speeds up the article production process. However, it's not that straightforward.

To note, I always compose my articles in English, despite the convenience of writing in French and relying on AI translation. It's a beneficial exercise for me, so I've chosen to stick with it.

While it's possible to directly request an article on a subject with minimal information, the subsequent adjustments can consume more time.

Once you have the foundation of your article, it's time to articulate your requirements.

I never use the same prompt, but the underlying principle remains consistent: provide as much information as possible.

For instance, my previous article took a different approach, focusing on installation notes I documented during the setup process. The initial prompt was:

Next message, I'll give you some notes I took while installing an application.
Make a tech article about it.
Give me a Markdown formatted text.
Lines starting with # are shell instructions.
Write a long article.

However, this wasn't sufficient, and additional instructions were needed, such as:

- Change "we" to "I."
- Replace my email address, users, and passwords with placeholders.

The next step involves reviewing what the AI generated. Evaluate what aligns with your preferences and what doesn't. Now begins the time-consuming phase.

Adapting the generated text to match your expectations is necessary. You can request changes to certain parts or rewrite them yourself. Adjustments may include adding new paragraphs or changing the order of information. Keep in mind that the AI may have an output limit, requiring you to divide your text into multiple parts, as I did for this very article.

Each time I rewrite a segment, I request a correction for the revised text. Sometimes, this process is iterated 2 or 3 times as new responses may inspire additional ideas.

Once the foundational article is written, I dedicate one to two hours working with ChatGPT to achieve the desired level of quality.

To write this article, I had to split the text to achieve the desired output. I utilized the following prompt:

I will give you a tech article I wrote concerning generative AI, paragraph by paragraph.
Improve the English and the overall quality and readability of the text, but use simple English.
Output the text in markdown format.
Text:
[My Text]

In the end, you'll still need to review the pagination and incorporate images along with their descriptions, which is a time consuming task as well.!

Image Generation

You've likely come across numerous instances in the news. The thumbnails for my last seven articles were generated using Bing Image Creator. Once again, it proves to be a formidable aid, saving me days that would otherwise be spent creating subpar images on my own. However, mastering image AI poses its own challenges.
Nonetheless, the results remain remarkably impressive!

I tried an AI inception here, asking to Bard to generate a prompt for Bing Image Creator.

Result with the original text :

Ok, I wrote an blog post about generative AI's. I want a thumbnail image describing it.
I love the belgian comic books, like Spirou, Tintin, Blake and Mortimer, so I'd like this kind of style.
It has to be a man in front of a computer. 
There should be something indicating the computer's ability to talk to hint at AI's. 
I'd like the background to be easily extensible as I will have to add a text to the right of the image. 
A simple continued gradient would be perfect

Bing Image Creator - The style is present, but the bubble comes from the man
The background rule is not effective

Result with the generated prompt :

Create an image in the style of Belgian comic books like Spirou, Tintin, and Blake and Mortimer. 
The image should feature a man sitting at a computer, with a speech bubble coming from the computer. 
The background should be a simple continued gradient that can be easily extended to the right to accommodate text.

Bing Image Creator - The result is cleaner, yet the bubble is empty and still coming from the man, except for the third image. Background rule is effective.

There is an improvement with the prompt generated by Bard. However, it's not flawless yet. One might contemplate whether the results would be significantly different with a second generation.

Conclusion

Generative AI opens up numerous possibilities. As demonstrated in this article, its applications span various aspects of daily life. I haven't even scratched the surface of its potential in professional environments or automation. I hope this article piques your curiosity. Feel free to share your thoughts in the comments!

Informations

For the thumbnail image, I used Bing Image Creator with the following prompt : "man working on a computer, speaking many languages, french cartoon" and reworked it using Canva.